In this paper, we evaluate the performance achieved by DWRR and WFQ, the widespread implementations of frame-based and sorted-priority schedulers respectively, in a high speed DiffServ environment. Using deterministic upper bounds of the end-to-end delay, we evaluate the schedulers behavior for a wide range of bandwidth configurations. Our analysis highlights the drawbacks of using frame-based scheduling with respect to sorted-priority when dealing with multiple classes of service. As a consequence, we propose a configuration of scheduler weights, which allows to contain the loss in performance of frame-based schemes, while minimizing the end-to-end delay for premium classes. Extensive simulations confirm our theoretical propositions and show the effectiveness of our method in satisfying hard quality of service requirements even in highly congested networks.